JP2011175362A - Information processing apparatus, importance level calculation method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To flexibly evaluate importance levels about various combinations between attributes from attributes of contents. <P>SOLUTION: The information processing apparatus includes: a memory holding an attribute table storing an attribute value imparted to each content about the plurality of contents; and an importance level calculation unit calculating the importance levels of other one or more attributes to a prescribed attribute of the content by use of the attribute values stored in the attribute table. The importance level calculation unit calculates the importance level by use of a determination table wherein the prescribed attribute is set as a determination attribute and wherein the one or more attributes are set as condition attributes. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an information processing apparatus, an importance calculation method, and a program.

  In recent years, with the development of information communication technology, various contents such as music, video, electronic books, news articles, product information or event information are provided to users via a network. It is not easy for individual users to search for information suitable for themselves from such a vast amount of content. Therefore, there are provided search services such as keyword search or genre search that assist the user in finding content, and recommendation services in which the system recommends content suitable for the user. For example, Patent Document 1 below describes a technique called collaborative filtering that is incorporated in many content recommendation services. Patent Document 2 below describes a method for determining a TV program to be recommended using matching based on a vector space method.

  In the content search service, content recommendation service, and other content-related services as described above, the key to the effectiveness of the service is the handling of content attributes. In general, there are various types of content attributes, such as attributes that are artificially given to content, attributes that are generated by analyzing content data, and attributes that are calculated based on user actions on content. There is. Today, not only the content provided to users has increased, but the types of content attributes that the service should handle are also increasing. Therefore, a technique that can efficiently extract important attributes for content classification, search, recommendation, and the like from content attributes has attracted attention. For example, Patent Document 3 below describes an example of a technique for extracting an important attribute from a plurality of attributes for data classified into categories.

JP 2007-323315 A Japanese Patent Laid-Open No. 2007-200339 JP 9-325969 A

  However, in the technique described in Patent Document 3, since importance is determined for each attribute, it is not possible to recognize the importance of a combination of attributes that are related to each other. For example, “genre” and “age” are examples of attributes of music content. The attribute value for “genre” is “rock”, “pop”, “classic”, or the like. The attribute value for “age” is “1970s”, “1980s”, “1990s”, or the like. In such a case, even if the “genre” (= “rock”, for example) is the same, the “rock” of which “age” has a different meaning for the user. That is, in general, if the importance of various combinations of attributes among a large number of content attributes can be flexibly evaluated, the reason for recommendation at the time of content recommendation and the addition of a genre for content search are provided. It is expected that the importance can be used for various purposes.

  Therefore, the present invention aims to provide a new and improved information processing apparatus, importance calculation method, and program capable of flexibly evaluating the importance of various combinations of attributes among the attributes of content. Is.

  According to an embodiment of the present invention, for a plurality of contents, a content is stored using a storage unit that stores an attribute table that stores attribute values assigned to each content, and attribute values stored in the attribute table. An importance calculating unit that calculates the importance of one or more other attributes with respect to the predetermined attribute, wherein the importance calculating unit uses the predetermined attribute as the determination attribute, There is provided an information processing apparatus that calculates using a decision table having two or more attributes as condition attributes.

  According to such a configuration, the importance for the predetermined attribute can be calculated for various combinations using the one or more attributes as a set of condition attributes using the determination table.

  In addition, the importance calculation unit may calculate the importance based on the number of contents forming a positive region of the determination attribute for the determination table.

  In addition, the information processing apparatus extracts one or more important attributes used to generate information about the content to be presented to the user according to the importance calculated by the importance calculation unit. May be further provided.

  The information processing apparatus is a recommendation unit that selects content to be recommended to a user using an attribute value stored in the attribute table, and the one or more important attributes extracted by the extraction unit A recommendation unit that generates a reason for recommendation based on the information may be further provided.

  Further, the recommendation unit further stores a score for each content calculated for content selection in the attribute table, and the extraction unit calculates the importance calculated by the importance calculation unit using the score as a determination attribute. Depending on the degree, the one or more important attributes may be extracted.

  Further, the attribute table further stores attribute values of feedback attributes given based on user feedback for each content, and the recommendation unit uses the attribute values of the feedback attributes to recommend content to the user The extraction unit may extract the one or more important attributes according to the importance calculated by the importance calculation unit using the feedback attribute as a determination attribute.

  The attribute table further stores an attribute value of a context attribute given based on a user action status for each content, and the recommendation unit should recommend to the user using the attribute value of the context attribute The content may be selected, and the extraction unit may extract the one or more important attributes according to the importance calculated by the importance calculation unit using the context attribute as a determination attribute.

  The attribute table may store an attribute value of an extended attribute obtained by analyzing the attribute value of the basic attribute in addition to the attribute value of the basic attribute given to each content.

  In addition, the extraction unit includes the first importance calculated by the importance calculation unit using the predetermined attribute as a determination attribute and one or more attributes included in the extended attribute as a condition attribute, and the extended attribute. The one or more important attributes according to the second importance calculated by the importance calculating unit using the attribute included in the basic attribute as a decision attribute and one or more attributes included in the basic attribute as a condition attribute May be extracted.

  Further, the information processing apparatus includes an analysis unit that calculates an attribute value of the extended attribute based on an attribute value of the basic attribute according to a probabilistic classification method based on PLSA (Probabilistic Latent Semantic Analysis) or LDA (Latent Dirichlet Allocation), May be further provided.

  The information processing apparatus may be a content list generation unit that generates a list of contents to be played back using the attribute values stored in the attribute table, and the one or more extracted by the extraction unit A content list generation unit that generates a title of the content list based on the important attribute, and the extraction unit uses the attribute used for generation of the content list by the content list generation unit as the determination attribute. The one or more important attributes may be extracted according to the importance calculated by the degree calculation unit.

  The information processing apparatus may be a content list generation unit that generates a list of contents to be played according to a user's specification, and the content list is based on the one or more important attributes extracted by the extraction unit A content list generation unit that generates a title of the image, and the extraction unit calculates the importance calculated by the importance calculation unit using an attribute whose attribute value is determined according to whether or not specified by the user as a determination attribute Depending on, the one or more important attributes may be extracted.

  Further, the storage unit further holds history data representing a history of user actions for each content, and the importance calculation unit uses the attribute value for the content included in the history data, for each user. The importance degree for one or more attributes may be calculated.

  In addition, the importance calculation unit may derive a positive region of the decision attribute in the decision table according to a Rough set theory.

  According to another embodiment of the present invention, in an information processing apparatus that uses a storage medium to store an attribute table that stores an attribute value assigned to each content for a plurality of contents, Calculating an importance level of one or more other attributes with respect to a predetermined attribute of the content using a stored attribute value, wherein the importance level is calculated based on the predetermined attribute value. Is a decision attribute, and an importance calculation method is provided that is calculated using a decision table having the one or more attributes as condition attributes.

  According to another embodiment of the present invention, for a plurality of contents, a computer that controls an information processing apparatus that holds an attribute table that stores attribute values assigned to each content using a storage medium, A program for functioning as an importance calculation unit that calculates importance of one or more other attributes with respect to a predetermined attribute of content using attribute values stored in the attribute table, The degree calculation unit is provided with a program for calculating the importance using a determination table in which the predetermined attribute is the determination attribute and the one or more attributes are condition attributes.

  As described above, according to the information processing apparatus, the importance calculation method, and the program according to the present invention, the importance of various combinations of attributes can be flexibly evaluated from the attributes of the content.

It is a block diagram which shows the structure of the information processing apparatus which concerns on one Embodiment. It is explanatory drawing for demonstrating the outline | summary of the attribute table which concerns on one Embodiment. It is explanatory drawing for demonstrating an example of the metadata of the basic attributes memorize | stored by the attribute table. It is explanatory drawing for demonstrating an example of the context data among the basic attributes memorize | stored by the attribute table. It is explanatory drawing for demonstrating the outline | summary of the extended attribute memorize | stored by an attribute table. It is explanatory drawing for demonstrating a specific example of the extended attribute shown in FIG. It is explanatory drawing for demonstrating an example of the evaluation attribute memorize | stored by the attribute table. It is explanatory drawing for demonstrating the various concepts relevant to a decision table. It is explanatory drawing for demonstrating calculation of the importance based on a determination table. It is explanatory drawing for demonstrating an example of the importance calculated based on a determination table. It is explanatory drawing for demonstrating the other example of the importance calculated based on a determination table. It is explanatory drawing for demonstrating the stepwise extraction of an important attribute. It is explanatory drawing for demonstrating an example of the screen as which a reason for recommendation is shown. It is explanatory drawing for demonstrating the other example of the screen as which the reason for recommendation is shown. It is a flowchart which shows an example of the flow of the pre-processing by the information processing apparatus which concerns on one Embodiment. It is a flowchart which shows an example of the flow of the recommendation process by the information processing apparatus which concerns on one Embodiment. It is a block diagram which shows the structure of the information processing apparatus which concerns on one modification. It is explanatory drawing for demonstrating the personalization which concerns on another modification. It is a block diagram which shows an example of a hardware configuration.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

Further, the “DETAILED DESCRIPTION OF THE INVENTION” will be described in the following order.
1. Explanation of terms 2. Configuration example of information processing apparatus according to one embodiment 2-1. Attribute table 2-2. Storage unit 2-3. Analysis unit 2-4. User interface controller 2-5. Recommendation section 2-6. Importance calculation unit 2-7. Extraction unit 2-8. Screen example 3. Flow of processing according to one embodiment 3-1. Pre-processing 3-2. Recommendation process Modified example 4-1. Provision of playlist 4-2. Personalization 5. 5. Hardware configuration example Summary

<1. Explanation of terms>
First, explanations of main terms used in this specification will be described below.
Probabilistic classification: A method for classifying elements of a set such as content or text into subsets. In the probabilistic classification method, one content or text can belong to a plurality of subsets with probability.
Latent topic: a concept that corresponds to an individual subset in a probabilistic taxonomy and potentially contributes to the occurrence of each content or text. It can be thought of as representing the field or topic of an individual subset.
PLSA (Probabilistic Latent Semantic Analysis): A probabilistic classification method. Provides a probabilistic generation model with latent topics and is widely used in the field of text classification.
LDA (Latent Dirichlet Allocation): One of the probabilistic classification methods developed from PLSA. Provides a probabilistic generation model with latent topics and is widely used in the field of text classification.
-Topic set: A set of content or text classified as belonging to a potential topic with attribution probability-Rough set theory: To analyze a decision table with uncertainty due to indiscernibility The theory underlying
Crisp-Rough set theory: A kind of Rough set theory. A technique for finding a reasonable approximation of an object by selecting attributes within a range that can identify the set of objects at a certain level.
Fuzzy-Rough set theory: A type of Rough set theory developed from Crisp-Rough set theory. A technique that enables the handling of attributes by continuous values and the description of the target by extending the target attribute from the nominal attribute to the numeric attribute.
Recommendation engine: A system module that recommends content based on the user's preferences or actions on the user's content. Recommendation engines based on various recommendation algorithms such as CF or CBF have already been put into practical use.
CF (Collaborative Filtering): collaborative filtering. A kind of recommendation algorithm. A method of accumulating preference data of a plurality of users and making recommendations based on data relating to other users who have similar preferences to a certain user.
CBF (Content Based Filtering): Content-based filtering. A kind of recommendation algorithm. A method for recommending content based on the similarity of content attribute data.
・ Reason for recommendation: Explanation regarding the basis of recommendation presented to the user when the recommendation engine recommends content.

<2. Configuration Example of Information Processing Device According to One Embodiment>
The information processing apparatus 100 according to an embodiment of the present invention typically holds an attribute table that stores attribute values for a plurality of contents, and has an importance level indicating the degree of contribution of the contents to a predetermined attribute. A device that calculates one or more other attributes. In the present embodiment, the information processing apparatus 100 further has a recommendation function. Then, the information processing apparatus 100 selects content to be recommended to the user, generates a reason for recommendation according to the calculated importance, and presents the information about the content to be recommended and the reason for recommendation to the user.

  FIG. 1 is a block diagram showing a configuration of an information processing apparatus 100 according to an embodiment of the present invention. Referring to FIG. 1, the information processing apparatus 100 includes a storage unit 110, an analysis unit 120, a user interface (UI) control unit 130, a recommendation unit 140, an importance calculation unit 150, and an extraction unit 160. The information processing apparatus 100 may be, for example, a content player that can reproduce content stored in the storage unit 110 via the UI control unit 130. Instead, the information processing apparatus 100 may be, for example, a content server that provides content data to the terminal device via the UI control unit 130 via a network. More generally, the information processing apparatus 100 may be any type of apparatus such as a high-performance computer, a PC (Personal Computer), a digital home appliance, a game machine, an AV player, or a smartphone.

[2-1. Attribute table]
First, an attribute table held by the storage unit 110 of the information processing apparatus 100 according to the present embodiment will be described. The attribute table is a table that stores one or more attribute values assigned to each content for a plurality of contents.

  FIG. 2 is an explanatory diagram for explaining an overview of the attribute table according to the present embodiment. Referring to FIG. 2, the attribute table is roughly divided into three types of attribute categories: basic attributes, extended attributes, and evaluation attributes. Basic attributes are further classified into two categories, metadata and context data. Each of these basic attributes (metadata), basic attributes (context data), extended attributes, and evaluation attributes includes one or more attribute items (individual attributes). An attribute value is assigned to each content item for each content item. Hereinafter, more specific examples of attributes will be described for each attribute category shown in FIG.

(1) Basic Attributes: Metadata FIG. 3 is an explanatory diagram for explaining an example of metadata among basic attributes stored by the attribute table. Referring to FIG. 3, the metadata includes five types of attributes of “genre”, “age”, “mood”, “keyword”, and “artist”. These attributes are expressed as numerical attributes in the example of FIG. For example, for “genre”, a numerical value indicating whether each content belongs to “G1: Rock”, “G2: Pop”, and other genres is given as an attribute value. For example, the attribute value for “genre” of the content C1 is (G1, G2,...) = (1.0, 0.0,...). Further, the attribute value for the “genre” of the content C2 is (G1, G2,...) = (0.0, 1.0,...). Similarly, for “age”, “E1: '70s (1970s)”, “E2:' 80s (1980s)”, and other numerical values indicating whether each content belongs to the other age are given as attribute values. It has been. The attribute value for the “age” of the content C1 is (E1, E2,...) = (1.0, 0.0,...). Further, the attribute value for the “age” of the content C2 is (E1, E2,...) = (0.0, 1.0,...). One content may belong to a plurality of “genres” or “age” with weights.

  The “mood” is an attribute given by analyzing the feature amount of the audio signal of the music content using, for example, the technique described in Japanese Patent Application Laid-Open No. 2007-207218. For example, a numerical value indicating whether or not each content has an impression such as “M1: brightness” or “M2: fun” is given to each content as an attribute value for “mood”. The “keyword” is an attribute given by, for example, text analysis of a review sentence supplied from a service provider or a user in association with each content. For example, the appearance frequency for each word such as a noun or adjective included in the review sentence is given to each content as an attribute value for “keyword”. “Artist” is an attribute assigned to each person name associated with the music content, for example. For example, a numerical value indicating whether or not there is a relationship with a person such as a composer, a lyricist, a singer or a co-star is given to each content as an attribute value for “artist”.

  Note that the attributes shown in FIG. 3 can be expressed not in the form of numerical attributes but also in the form of nominal attributes. For example, according to the format of the nominal attribute, “genre” = “G1: Rock”, “age” = “E1: '70s” for the content C1, and “genre” = “G2: Pop”, “age” for the content C2. = “E2: '80s” (“attribute” = “attribute value”).

  Typically, these metadata can be given in advance to each content independently of actions such as browsing, viewing, purchasing, etc. (hereinafter referred to as browsing) by the user, or recommendation processing described later. It is data.

(2) Basic Attributes: Context FIG. 4 is an explanatory diagram for explaining an example of context data among basic attributes stored by the attribute table. The context data is typically data that is given based on the state of user action for each content. Referring to FIG. 4, the context data includes two types of attributes “time zone” and “location”. For example, “time zone” represents the number of times each content is browsed in “T1: 10-12 (hours)”, “T2: 12-14 (hours)”, and other time zones. For example, the attribute value for the “time zone” of the content C1 is (T1, T2,...) = (2, 10,...). Also, the attribute value for the “time zone” of the content C2 is (T1, T2,...) = (3, 1,...). Similarly, “location” represents the number of times each content is browsed in “P1: Tokyo”, “P2: Osaka” and other locations. The attribute value for the “location” of the content C1 is (P1, P2,...) = (8, 4,...). Further, the attribute value for “location” of the content C2 is (P1, P2,...) = (0, 4,...). In addition to the attributes illustrated in FIG. 4, the context data may include any attribute that can be acquired in association with the user action, such as a day of the week, a date, a country (location), a user gender, or an age group. Further, instead of the geographical location such as “Tokyo” or “Osaka” illustrated in FIG. 4, attribute values for “location” such as “home”, “office”, or “in the car” are included in the context data. May be.

(3) Extended Attributes FIG. 5 is an explanatory diagram for explaining an outline of extended attributes stored by the attribute table. The extended attribute is an attribute obtained by analyzing the attribute value of the basic attribute given to each content. In the present embodiment, the attribute table has extended attributes that are divided into two layers, a review sentence layer and a person layer. Among these, the review sentence layer represents the attribution probability of each content for each of the latent topics X1,..., Xm obtained by analyzing the attribute value of “keyword” among the basic attributes according to the probabilistic classification method. On the other hand, the person layer represents the attribution probability of each content for each of the latent topics Y1,..., Ym, obtained by analyzing the attribute value of “artist” among the basic attributes according to the probabilistic classification method.

  FIG. 6 is an explanatory diagram for further explaining a specific example of the extended attribute shown in FIG. The upper part of FIG. 6 shows the attribute values (frequency of each keyword in the review sentence) of the keywords (K1, K2, K3, K4, K5,...) For each of the contents C1, C2, and C3. The keywords K1, K2, K3, K4, and K5 are “scandal”, “album”, “popularity”, “record”, and “ability”, respectively. For the content C1, the keywords (K1, K2, K3, K4, K5,...) = (1, 0, 2, 0, 1,...). For the content C2, the keywords (K1, K2, K3, K4, K5,...) = (0, 1, 0, 2, 1,...). For the content C3, the keyword (K1, K2, K3, K4, K5,...) = (3, 0, 1, 1, 0,...). According to the probabilistic taxonomy modeled by PLSA or LDA, these keywords appear in the review text of each content due to the contribution of potential topics to which each content potentially belongs. Conversely, if the attribute values of the keywords (K1, K2, K3, K4, K5,...) Are analyzed according to the probabilistic classification method by PLSA or LDA, the attribution probability for each potential topic of each content can be calculated.

  More specifically, when the occurrence probability of the keyword w in the review sentence d of each content is p (w | d), the occurrence probability p (w | d) is expressed by Expression (1).

In Expression (1), x _i is a latent topic, p (w | x _i ) is the probability of occurrence of the word w for the latent topic x _i , and p (x _i | d) is the topic (of the review sentence d) of each content Distribution. Note that the number of latent topics x _i is set to an appropriate value (for example, 16) in advance according to the dimension of the data space to be analyzed.

  In the lower part of FIG. 6, latent topics (X1, X2,..., Xn) obtained by analyzing attribute values of keywords (K1, K2, K3, K4, K5,...) According to a probabilistic classification method by PLSA or LDA. An example of the probability of belonging to each) is shown. For example, for the content C1, the attribution probability (X1, X2,..., Xn) = (0.4, 0.1,..., 0.3). For the content C2, the attribution probability (X1, X2,..., Xn) = (0.1, 0.2,..., 0.1). For the content C3, the attribution probability (X1, X2,..., Xn) = (0.6, 0.1,..., 0.1). In this embodiment, the attribution probability for each potential topic of each content is stored in the attribute table as an attribute value of one layer of extended attributes. When calculating the extended attribute according to the probabilistic classification method, the attribute value of the basic attribute is normalized within the range of one attribute item (for example, an individual keyword) (with the maximum value set to 1). It is preferred to apply a probabilistic classification method.

(4) Evaluation Attributes FIG. 7 is an explanatory diagram for explaining an example of evaluation attributes stored by the attribute table. The evaluation attribute is typically an attribute representing evaluation by a recommendation engine or a user for each content. Referring to FIG. 7, evaluation attributes are classified into two categories: algorithm score and user feedback (FB).

  Among these, the algorithm score includes scores S1, S2,... For each content calculated by one or more recommendation algorithms. For example, the score S1 may be a score calculated by CF (collaborative filtering), and the score S2 may be a score calculated by CBF (content-based filtering). In this case, for example, the score S1 for the content C1 corresponds to the degree of similarity between the user preference of the user to be recommended and the user preference of another user who has browsed the content C1. The score S2 for the content C1 corresponds to the degree of similarity between the metadata of the content browsed by the user to be recommended and the metadata of the content C1.

  On the other hand, the user FB includes attribute values given based on feedback from the user with respect to each content. The attribute value of the user FB represents, for example, binary data of “Y: likes” or “N: dislikes” or a score in multi-level (for example, 5-level) evaluation. Instead, the user feedback may represent the number of actions such as browsing by the user, for example.

  Next, the operation of each unit of the information processing apparatus 100 illustrated in FIG. 1 will be described in order.

[2-2. Storage unit]
The storage unit 110 is configured using a storage medium such as a hard disk or a semiconductor memory, and holds the attribute table described with reference to FIGS. The attribute table stores attribute values of the attribute items of the basic attribute, the extended attribute, and the evaluation attribute described above. The storage unit 110 inputs and outputs these attribute values with each unit of the information processing apparatus 100. Furthermore, the storage unit 110 may store the content data itself. For example, when the information processing apparatus 100 is a music player, audio data of music content may be stored in the storage unit 110.

[2-3. Analysis Department]
The analysis unit 120 calculates the attribute value of the extended attribute based on the attribute value of the basic attribute of the attribute table according to the probabilistic classification method by PLSA or LDA. For example, the analysis unit 120 calculates the attribute value of the review sentence layer of the extended attribute by analyzing the attribute value related to the keyword among the attribute values of the basic attribute according to the probabilistic classification method. More specifically, the attribute value of the review sentence layer of the extended attribute is, for example, the topic distribution of each content that derives the attribute value (appearance frequency) related to the keyword by the product sum with the occurrence probability of the keyword for each latent topic. (See equation (1)). For example, the analysis unit 120 calculates the attribute value of the extended attribute person layer by analyzing the attribute value related to the artist among the attribute values of the basic attribute according to the probabilistic classification method. Then, the analysis unit 120 stores the calculated attribute value of the extended attribute in the attribute table. The analysis according to the probabilistic classification method by the analysis unit 120 is performed periodically, for example, when a certain number of contents are accumulated in the content database (for example, the storage unit 110), or every month or every year. Can be implemented automatically.

  The PLSA is described in detail in “Probabilistic latent semantic indexing” by Thomas Hofmann (1999, Proceedings of the 22nd annual international ACM SIGIR conference on Research and development in information retrieval). LDA is described in detail in “Latent Dirichlet Allocation” (2003, Journal of Machine Learning Research, Volume 3) by David M. Blei, Andrew Y. Ng, Michael I. Jordan.

[2-4. User interface control unit]
The UI control unit 130 controls a user interface between the information processing apparatus 100 and the user. Such a user interface typically includes a screen interface displayed by a display device and an input interface such as a mouse, a keyboard, a touch panel, or a keypad. Further, an interface for content reproduction such as an audio output circuit or a video output circuit may be included in the user interface. For example, these user interfaces may be mounted on the information processing apparatus 100, or instead, may be mounted on a terminal device connected to the information processing apparatus 100 via a network.

  More specifically, the UI control unit 130 updates the context data for the content according to, for example, an action by the user for the certain content. That is, for example, assume that the user browses the content C1 in Tokyo at 11:00 am. In that case, the UI control unit 130 updates the attribute values of the time zone “T1: 10 to 12 (hours)” and the location “P1: Tokyo” for the content C1. For example, the UI control unit 130 can recognize the viewing location of the content by the user based on the IP address of the terminal device or using the GPS (Global Positioning System).

  For example, the UI control unit 130 acquires the user feedback value illustrated in FIG. 7 for a certain content via the input interface, and stores the acquired value in the attribute table. In addition, the UI control unit 130 causes the display device to display a recommendation screen generated by the recommendation unit 140 described later, for example.

  Furthermore, for example, when the information processing apparatus 100 or the terminal device includes a sensor that can detect the user state, the UI control unit 130 uses the user state acquired by the sensor as an attribute value of the context data. It may be stored in the attribute table. For example, it is conceivable to acquire the state of the user's smile using an imaging module having a smile sensor function, or to acquire the state of the user's pulse or respiration rate using a biological sensor.

[2-5. Recommendation section]
The recommendation unit 140 operates as a recommendation engine that selects content to be recommended to the user using the attribute values stored in the attribute table held by the storage unit 110. The recommendation unit 140 generates a reason for recommendation based on one or more important attributes extracted by the extraction unit 160 described later. And the recommendation part 140 produces | generates the recommendation screen which shows the information regarding the content which should be recommended, and a recommendation reason to a user, and displays the produced | generated recommendation screen on a display apparatus via UI control part 130. FIG.

  The selection of content by the recommendation unit 140 may be performed according to any known recommendation algorithm such as the above-described CF or CBF. For example, when the recommendation unit 140 uses CF as a recommendation algorithm, the recommendation unit 140 specifies another user having a preference similar to the user's preference. Next, the recommendation unit 140 extracts a group of contents browsed by the other user. At this time, the recommendation unit 140, for example, according to the similarity of the preference or the evaluation of each content by the other user, the score for each of the extracted group of content (the degree of suitability of each content to the user, or Calculate certainty about recommendation). The score for each content calculated here is stored in the attribute table. And the recommendation part 140 selects the content with a relatively high score as content which should be recommended to a user, for example.

  Further, for example, when the recommendation unit 140 uses CBF as a recommendation algorithm, the recommendation unit 140 extracts a group of contents whose metadata is similar to content viewed by the user. At this time, the recommendation unit 140 calculates a score for each of the extracted group of contents, for example, according to the similarity of the metadata. The score for each content calculated here is stored in the attribute table. And the recommendation part 140 selects the content with a relatively high score as content which should be recommended to a user, for example.

  For example, the recommendation unit 140 may select content to be recommended to the user using the attribute value for the user FB stored in the attribute table. For example, among the plurality of contents, a content having a good attribute value for the user FB can be preferentially selected as the content to be recommended. Further, for example, the recommendation unit 140 may select content to be recommended to the user by using the attribute value for the context data stored in the attribute table. For example, among a plurality of contents, content that has been browsed more frequently in the time zone where the recommendation process is being executed and the location of the user to be recommended is preferentially selected as the content to be recommended Can be done.

  When selecting the content to be recommended, the recommendation unit 140 causes the importance level calculation unit 150 to calculate the importance level for the predetermined attribute used for selecting the content for one or more attributes other than the predetermined attribute. In addition, the recommendation unit 140 causes the extraction unit 160 to extract one or more important attributes used to generate a recommendation reason to be presented to the user. Then, the recommendation unit 140 generates a recommendation reason using one or more important attributes extracted by the extraction unit 160, and displays a recommendation screen that displays information about the content to be recommended and the recommendation reason, and the UI control unit 130. Output to.

[2-6. Importance calculator]
The importance level calculation unit 150 uses the attribute values stored in the attribute table to calculate the importance level for the predetermined attribute of the content for one or more attributes other than the predetermined attribute of the content. In particular, in the present embodiment, the importance level calculation unit 150 uses a determination table in which the importance level for the one or more attributes is set as the determination attribute and the one or more attributes are condition attributes. To calculate. For example, as will be described below, the importance level calculation unit 150 can calculate the importance level based on the number of contents forming the positive region of the determination attribute for the determination table described above. The positive region of the decision attribute in the decision table can be derived according to Rough set theory.

(1) Various Concepts Related to Decision Table FIG. 8 is an explanatory diagram for explaining various concepts related to the decision table to which the Rough set theory is applied. The basic configuration of the decision table is shown in the upper part of FIG. The decision table illustrated in FIG. 8 is a two-dimensional table, in which the vertical axis represents the target total set U and the horizontal axis represents the attribute total set A. The entire target set U is a set of contents. On the other hand, the entire attribute set A is a set of attribute items. The entire attribute set A is a union of the condition attribute C and the decision attribute D. Both the condition attribute C and the decision attribute D include one or more attributes. The attribute value set V is a set of attribute values of nominal attributes that each attribute included in the attribute total set A can take. For example, when the attributes “genre” and “age” illustrated in FIG. 3 are members of the entire attribute set A, V = {{Rock, Jazz,...}, {'70s,' 80s,. }} Etc. The attribute value function ρ is a function that specifies an attribute value in response to one element (content) of the entire object set U and one element (attribute) of the entire attribute set A (ρ). : U × A → V).

  A concept set X is shown in the lower left of FIG. The concept set X is a set of contents belonging to the concept when the attribute value of a specific attribute is regarded as one concept. For example, when the attribute A can take a binary attribute value “Y” or “N”, the concept set X for the concept of the attribute A = “Y” Include as.

  In the lower right of FIG. 8, attribute subsets P (P⊆C) and Q (Q⊆D) are shown. The attribute subset P is a set of condition attributes used for calculating the importance of the condition attributes C. On the other hand, the attribute subset Q is a set of decision attributes used for calculation of importance among the decision attributes C.

(2) Definition Formula for Crisp Rough Set Table 1 shows a definition formula for the Crisp Rough set described using various concepts related to the decision table described above.

  Each item in Table 1 will be described with reference to FIG. In the upper part of FIG. 9, an example of a decision table including the contents C1 to C6 in the target entire set U is shown. The attribute subset P of the condition attributes in the decision table of FIG. 9 includes metadata “genre”, “age”, “keyword (“ world ”)”, and “mood (“ brightness ”)” among the basic attributes. . Here, from the viewpoint of simplicity of explanation, it is assumed that “genre”, “age”, and “mood (“ brightness ”)” are expressed in the form of nominal attributes. “The attribute subset Q of the decision attributes includes the user FB“ preference ”among the evaluation attributes.

  In Table 1, the expression of P-equivalence relationship represents a condition for content x and content y belonging to the target entire set U to have an equivalence relationship in the attribute subset P of the condition attributes. For example, in the determination table of FIG. 9, the attribute value of the condition attribute of the content C2 is equal to the attribute value of the condition attribute of the content C6, and the content C2 and the content C6 are in the same value relationship.

  The P-equivalence class represents a set of contents having an equivalence relation with a certain content x in the attribute subset P of the condition attributes. In the decision table of FIG. 9, the content C2 and the content C6 form one P-equivalence class.

  The U division by P-equivalence classes indicates that the target entire set U can be divided with each P-equivalence class as a subset. In the decision table of FIG. 9, the target entire set U can be divided by five subsets {C1}, {C2, C6}, {C3}, {C4}, and {C5}.

  The P-top approximation of the concept X represents a set of contents that are determined to be included in the concept set X based on the attribute values for the attribute subset P of the condition attribute. In the decision table of FIG. 9, contents C1, C2, C4, and C6 are contents that may be included in the concept set X of “preference” = “Y” of the decision attribute Q. Therefore, the upper approximation of the concept of “preference” = “Y” is the set {C1, C2, C4, C6}. The reason why the content C2 is included in the upper approximation is that the “favorite” of the content C6 that is equivalent to the content C2 is “Y”.

  The P-lower approximation of the concept X represents a set of contents that are determined to be necessarily included in the concept set X based on the attribute values for the attribute subset P of the condition attribute. In the decision table of FIG. 9, the contents that are inevitably determined to be included in the concept set X of “preference” = “Y” of the decision attribute Q are the contents C1 and C4. Therefore, the lower approximation of the concept of “preference” = “Y” is the set {C1, C4}. The reason why the content C6 is not included in the lower approximation is that the “favorite” of the content C2 having the same relationship with the content C6 is “N”.

  The U division by the Q-equivalence class indicates that the target entire set U can be divided by using each Q-equivalence class as a subset. In the decision table of FIG. 9, the target entire set U can be divided by two subsets {C1, C4, C6} and {C2, C3, C5} that are equivalence classes of the decision attribute Q.

  The positive region in P and Q represents the union over all equivalence classes of the P-lower approximation for the equivalence class of the decision attribute Q. In the decision table of FIG. 9, the lower approximation of the concept of “preference” = “Y” is the set {C1, C4}. The lower approximation of the concept of “preference” = “N” is the set {C3, C5}. Accordingly, the primary regions in the condition attribute P and the decision attribute Q in the decision table of FIG. 9 are the set {C1, C4}, {C3, C5}}.

(3) Definition Formula for Fuzzy Rough Set Table 2 shows a definition formula for the Fuzzy Rough set for the concepts related to the decision table described above. By using the Fuzzy Rough set instead of the Crisp Rough set, it is possible to handle not only the approximate set related to the nominal attribute but also the approximate set related to the numerical attribute. For example, in Table 2, (membership value of the attribute _{F i} in the subject x) fuzzy equivalence classes _mu Fi (x) is represented by consecutive values of 0.0 to 1.0.

  The Fuzzy Rough set theory is detailed in “Fuzzy-Rough Data Reduction with Ant Colony Optimization” by Richard Jensen and Qiang Shen (Fuzzy Sets and Systems, vol. 149, no. 1, pp. 5-20, 2005). Since it is described, its description is omitted here.

(4) Importance Calculation Formula The importance calculation formula for the Crisp Rough set by the importance calculation unit 150 is shown in Formula (2). γ _P (Q) is the importance of the condition attribute (set) P with respect to the decision attribute (set) Q.

  In Equation (2), card (K) represents the cardinality of the set K, that is, the original number included in the set K. That is, the denominator on the right side represents the density of the entire target set U. The numerator on the right side represents the density of the positive region for the condition attribute (set) P and the decision attribute (set) Q.

  In the example of FIG. 9, the density of the entire target set U is the number of contents C1 to C6, that is, six. The density of the positive region for the condition attribute (set) P and the decision attribute (set) Q is the density of the lower approximation (= 2) of “Preference” = “Y” and the density of the lower approximation of “Preference” = “N”. It corresponds to the sum of (= 2), and 2 + 2 = 4. Therefore, according to the formula (2), the combination of the condition attribute “genre”, “age”, “keyword (“ world ”)” and “mood (“ brightness ”)” with respect to the decision attribute (“preference”) is important. The degree is calculated as 4/6 = 0.67.

  Generally, the density of the positive region is determined when the attribute value of the selected condition attribute greatly contributes to the determination of the attribute value of the determination attribute, that is, the attribute value of the determination attribute is given by the attribute value of the condition attribute. A high value is obtained when prediction can be made with higher accuracy. On the other hand, if the attribute value of the decision attribute is not determined with a certain degree of accuracy (or necessity) even if the attribute value of the condition attribute is given, the density of the normal region is low. This means that a combination of condition attributes (condition attribute attribute subset P) in which the density of the normal region normalized by using the density of the entire object set U is arbitrarily selected as shown in Equation (2). It can be treated as an index (or an approximate quality index) of informativity for the attribute value of the determination attribute (attribute subset Q of the determination attribute). Therefore, for example, by calculating the importance for each combination using the evaluation attribute included in the attribute table as the determination attribute and the combination of any other attribute as the condition attribute, the score or the user FB as the basis for recommending the content can be obtained. It is possible to numerically evaluate what combination of attributes can be determined or predicted significantly.

  The importance calculation formula for the Fuzzy Rough set by the importance calculation unit 150 is as shown in Expression (3). The importance calculated in accordance with the expression (3) is also transmitted to the attribute value of the decision attribute (the attribute subset Q of the decision attribute) of the arbitrarily selected combination of the condition attributes (the attribute subset P of the condition attribute) It can be treated as an index of sex.

  The importance calculation unit 150 calculates the importance for the decision attribute Q according to the formula (2) or the formula (3) for various combinations of condition attributes (hereinafter referred to as condition attribute sets).

  FIG. 10 is an explanatory diagram for explaining an example of importance levels for various condition attribute sets calculated based on the determination table illustrated in the upper part of FIG. 9. Referring to FIG. 10, 15 condition attributes that can be selected from condition attributes C including four elements of “genre”, “age”, “keyword (“ world ”)”, and “mood (“ brightness ”)”. Sets AS1 to AS15 are shown (some of which are omitted). For example, the condition attribute set AS1 includes only “genre”. The condition attribute set AS2 includes only “age”. The condition attribute set AS5 includes “genre” and “age”. The condition attribute set AS15 includes all of “genre”, “age”, “keyword (“ world ”)”, and “mood (“ brightness ”)”.

  The importance calculation unit 150 uses, for example, one attribute (“preference” in the example of FIG. 10) as a decision attribute, and sets each condition attribute set AS1 to AS15 as an attribute subset of condition attributes. The degree is calculated according to formula (2) or formula (3). As a result, for example, it is assumed that the condition attribute set AS7 shows the highest importance value. In the example of FIG. 10, the importance of the condition attribute set AS7 is 0.55. In this case, the combination of “genre” and “mood (“ brightness ”)” included in the condition attribute set AS7 is the most significant attribute for determining or predicting the attribute value of the determination attribute “preference”. be able to.

  The importance calculation unit 150 outputs the calculation result of the importance for each condition attribute set to the extraction unit 160.

[2-7. Extraction unit]
The extraction unit 160 extracts one or more important attributes used to generate information about the content to be presented to the user according to the above-described importance calculated by the importance calculation unit 150. The information to be presented to the user regarding the content may be, for example, the reason for recommendation presented together with the information regarding the content to be recommended. Instead, the information about the content to be presented to the user may be, for example, the title of a content list (for example, a playlist that lists music content). The important attribute extraction processing by the extraction unit 160 mainly includes two patterns: (1) direct extraction and (2) stepwise extraction.

(1) Direct extraction In the case of direct extraction, the extraction unit 160 uses, for example, an attribute that is a basis for content recommendation as a decision attribute, and uses one or more attributes that can be used for generating a recommendation reason as a conditional attribute. The importance calculated by the importance calculation unit 150 is used for extracting important attributes.

  For example, it is assumed that when the recommendation unit 140 selects content to be recommended using an attribute value for the user FB, the reason for recommendation is generated using content metadata. In this case, for example, as illustrated in FIG. 10, the importance calculation unit 150 performs user FB (condition attribute set) on various combinations (condition attribute sets) of content metadata (“genre”, “age”, etc.). The degree of importance is calculated when “preference”) is used as a decision attribute. Then, the extraction unit 160 extracts, for example, one or more attributes included in the condition attribute set having the highest calculated importance as the important attributes. That is, in the example of FIG. 10, “genre” and “mood (“ brightness ”)” can be extracted as important attributes.

  FIG. 11 is an explanatory diagram for describing another example of the importance calculated based on the determination table. In relation to the example of FIG. 11, for example, a situation is assumed in which content with a history of browsing or the like in a time zone (for example, T1) in which recommendation processing is executed can be selected as content to be recommended. In this case, for example, the importance calculation unit 150 uses context data (for example, “time zone: T1”) for various combinations (condition attribute sets) of attributes of content metadata (“genre”, “age”, etc.). The degree of importance is calculated when the is a decision attribute. Then, the extraction unit 160 extracts, for example, one or more attributes included in the condition attribute set having the highest calculated importance as the important attributes. In the example of FIG. 11, since the condition attribute set AS5 including “genre” and “age” has the highest importance (0.65), “genre” and “age” can be extracted as important attributes.

  Note that the extraction unit 160 does not extract the attribute included in the calculated condition attribute set having the highest importance level as an important attribute, but extracts one or more condition attribute sets that indicate an importance level exceeding a preset threshold. The included attributes may be extracted as important attributes. In addition, the extraction unit 160 may extract N (N is preset) attributes as important attributes in descending order of the calculated importance.

(2) Stepwise extraction In the case of stepwise extraction, the extraction unit 160 uses, for example, an attribute that is a basis for content recommendation as a decision attribute, and one or more attributes included in an extended attribute as important as a condition attribute. Calculated by the importance calculation unit 150 using the first importance calculated by the degree calculation unit 150 and the attribute included in the extended attribute as a decision attribute and one or more attributes that can be used for generating a recommendation reason as a condition attribute The second importance degree to be used is used for extraction of important attributes.

FIG. 12 is an explanatory diagram for explaining the stepwise extraction of important attributes by the extraction unit 160. In the upper part of FIG. 12, the score of the recommendation algorithm that is the basis of content recommendation is the determination attribute Q1, and any combination P1 of the latent topics X1 to Xn included in the review sentence layer of the extended attribute is the condition attribute. The calculation unit 150 calculates the first importance γ _P1 (Q1) ( _P1算出 C1). Here, for example, it is assumed that the first importance γ _P1 (Q1) is the maximum when P1 = {X1, X2}.

In the lower part of FIG. 12, P1 = {X1, X2} is a decision attribute Q2, and an arbitrary combination P2 of keywords K1, K2, K3,. 150, the second importance γ _P2 (Q2) is calculated. From the calculation results of the first importance level and the second importance level, the extraction unit 160, for example, sets the condition attribute set P2 (Q2 = P1 = {X1, Q2) having the highest second importance level γ _P2 (Q2). X2}) are extracted as important attributes.

  Note that the stepwise extraction method is not limited to such an example. The method described with reference to FIG. 12 uses the first importance to identify a set of extended attributes (latent topics in the review sentence layer) that most contributes to the score of the recommendation algorithm, and then to the specified set of extended attributes. This is a method of extracting a set of contributing basic attributes (keywords) using the second importance. Instead, for example, the extraction unit 160 multiplies the first importance level and the second importance level with respect to the common extended attribute set, and adds the attributes included in the basic attribute set whose multiplication result is larger. It may be extracted as an important attribute.

  Regardless of direct extraction or stepwise extraction, the combination of the condition attribute and the determination attribute is not limited to the above-described example. The extraction unit 160 outputs the important attributes extracted in this way to the recommendation unit 140.

[2-8. Screen example]
13 and 14 show examples of screens on which the reasons for recommendation are presented. Referring to FIG. 13, an example recommendation screen 132 a generated by the recommendation unit 140 and displayed on the display device via the UI control unit 130 is shown. The recommendation screen 132a includes a jacket image 134a of music content recommended by the user, a description sentence 136a of the content, and a recommendation reason 138a. In the recommendation reason 138a, character strings representing the important attributes extracted by the extraction unit 160 are listed. By displaying such a recommendation reason 138a, it is possible to convince the user of the reason for recommendation for the recommended content. Also, the user can more easily determine an action (viewing, purchasing, ignoring, etc.) for the recommended content.

  Referring to FIG. 14, a recommendation screen 132 b as another example generated by the recommendation unit 140 and displayed on the display device via the UI control unit 130 is shown. The recommendation screen 132b includes a jacket image 134b of music content recommended by the user, a description 136b of the content, and a reason for recommendation 138b. In the recommendation reason 138b, a sentence of the reason for recommendation generated by using the important attribute extracted by the extraction unit 160 is shown. Such a recommendation reason 138b can also convince the user of the reason for recommendation for the recommended content, and can more easily determine the user's action for the recommended content.

<3. Flow of processing according to one embodiment>
Next, the flow of processing by the information processing apparatus 100 according to the present embodiment will be described with reference to FIGS. 15 and 16.

[3-1. Pre-processing]
FIG. 15 is a flowchart illustrating an example of the flow of pre-processing by the information processing apparatus 100 according to the present embodiment. The pre-processing is processing that can be performed before the content recommendation processing. The pre-processing can be executed, for example, when a certain number of contents are accumulated or periodically every predetermined period.

  Referring to FIG. 15, first, the basic attribute of content stored in the attribute table of the storage unit 110 is acquired by the analysis unit 120 (step S <b> 102). Next, the analysis unit 120 calculates the attribute value of the extended attribute based on the attribute value of the basic attribute according to the probabilistic classification method by PLSA or LDA (step S104). The analysis unit 120 stores the attribute value of the extended attribute calculated here in the attribute table.

  Next, the importance level calculation unit 150 determines whether or not stepwise extraction of important attributes is performed in the recommendation process by the recommendation unit 140 (step S106). Here, if the important attributes are not extracted stepwise, step S108 is skipped. When the important attributes are stepwise extracted, the importance degree calculation unit 150 uses the extended attribute calculated by the analysis unit 120 as the decision attribute and the basic attribute as the condition attribute (in the stepwise extraction). The second importance) is calculated (step S108). For example, the importance calculation unit 150 stores the importance calculated here in the storage unit 110 for later recommendation processing.

  In addition to such pre-processing, for example, when an action such as browsing of content is performed by the user, processing such as updating of context data and registration of the user FB by the UI control unit 130 may be performed. Further, recalculation of extended attributes may be performed accordingly.

[3-2. Recommendation process]
FIG. 16 is a flowchart illustrating an example of a flow of recommendation processing by the information processing apparatus 100 according to the present embodiment. The recommendation process is a process that can be performed in response to a request from a user or a terminal device, for example.

  Referring to FIG. 16, first, the recommendation unit 140 selects content to be recommended to the user using the attribute values stored in the attribute table held by the storage unit 110 (step S202). Next, the recommendation unit 140 stores the score output by the recommendation algorithm in the attribute table as the attribute value of the evaluation attribute (step S204). Next, it is determined by the importance level calculation unit 150 whether or not the stepwise extraction of important attributes is performed (step S206). Here, when the important attribute is extracted stepwise, the process proceeds to step S208. On the other hand, if the important attributes are not extracted stepwise, the process proceeds to step S212.

  In step S208, the importance calculation unit 150, for example, the importance (stepwise extraction) when the score (or other attribute such as the user FB) output by the recommendation unit 140 is the determination attribute and the extended attribute is the condition attribute. (First importance level) is calculated (step S208). Next, the extraction unit 160 extracts important attributes according to the first importance level and the second importance level (step S210).

  On the other hand, in step S212, the importance calculation unit 150 uses, for example, the importance (directly) when the score (or another attribute such as the user FB) output by the recommendation unit 140 is the determination attribute and the basic attribute is the condition attribute. The importance in the extraction is calculated (step S212). Next, the extraction unit 160 extracts an important attribute according to the importance calculated by the importance calculation unit 150 (step S214).

  Next, the recommendation unit 140 generates a recommendation screen including a recommendation reason to be presented to the user based on the important attributes extracted by the extraction unit 160 (step S216). Next, the UI control unit 130 causes the display device to display the recommendation screen generated by the recommendation unit 140 (step S218). Thereafter, when the content to be recommended is changed, the processing returns to step S202, and the above-described processing is repeated. On the other hand, when there is no new content to be recommended, the recommendation process ends.

  Note that the flowchart shown in FIG. 16 shows an example in which either one of the important attribute direct extraction or stepwise extraction is performed for one content. However, the important attribute direct extraction or stepwise extraction is performed for one content. Both may be performed. For example, the information processing apparatus 100 directly extracts basic attributes (“genre”, “age”, “mood”, etc.) with a small original attribute value set, and basic attributes (“keyword” with a large attribute value set) ”,“ Artist ”, etc.) may be extracted step by step using extended attributes. The attribute space dimension of the extended attribute is generally lower than the attribute space dimension of the basic attribute (typically equivalent to the number of potential topics), so the extended attribute should be used as a conditional attribute in the middle Thus, the calculation cost of the importance calculation process during the recommendation process can be suppressed. Further, as an advantage of the probabilistic classification method, it is possible to reflect the relationship between potential keywords or artists (for example, synonymous relationships) in the calculation result of the importance.

<4. Modification>
[4-1. Provision of playlist]
FIG. 17 is a block diagram illustrating a configuration of an information processing apparatus 200 according to a modification of the present embodiment. Referring to FIG. 17, the information processing apparatus 200 includes a storage unit 110, an analysis unit 120, a user interface (UI) control unit 130, a content list generation unit 240, an importance calculation unit 250, and an extraction unit 260. In the information processing apparatus 200, a content list generation unit 240 is provided instead of the recommendation unit 140 illustrated in FIG.

  The content list generation unit 240 generates a list of contents to be played back by the UI control unit 130 in accordance with a user designation or using attribute values stored in the attribute table. In addition, the content list generation unit 240 generates a title of the content list based on one or more important attributes extracted by the extraction unit 260. Then, the content list generation unit 240 presents the content list with the title to the user via the UI control unit 130. The content list generated by the content list generation unit 240 may be, for example, a playlist that lists music content or video content to be reproduced.

  The content list generated by the content list generation unit 240 includes, for example, a group of contents specified by the user. When the user specifies content to be included in the content list, the content list generation unit 240, for example, as an evaluation attribute of the attribute table, an attribute value according to the presence / absence of designation by the user (for example, “1” if designated) If there is no designation, a score having “0”) is stored. On the other hand, the content list generation unit 240 stores the score output by the recommendation algorithm in the attribute table when one of the above-described recommendation algorithms is used to select content to be included in the content list.

  Further, for example, the content list generation unit 240 may select content to be included in the content list using the attribute value for the user FB stored in the attribute table. In addition, for example, the content list generation unit 240 may select content to be included in the content list using the attribute value for the context data stored in the attribute table. For example, content that has been played more frequently in a specific time zone can be preferentially selected as content to be included in the content list.

  When the content list is generated, the content list generation unit 240 calculates the importance for the predetermined attribute used for generating the content list to the importance calculation unit 250 for one or more attributes other than the predetermined attribute. Let In addition, the content list generation unit 240 causes the extraction unit 260 to extract one or more important attributes used for generating the title of the content list. Then, the content list generation unit 240 generates a title of the content list using one or more important attributes extracted by the extraction unit 260, and sends the content list with the title to the user via the UI control unit 130. Present.

  The importance level calculation unit 250 uses the attribute values stored in the attribute table to calculate the importance level for the predetermined attribute of the content for one or more attributes other than the predetermined attribute of the content. For example, the importance level calculation unit 250 may calculate the importance level using a determination table in which an evaluation attribute whose attribute value is determined according to whether or not a user designates is used as a determination attribute. Further, for example, the importance level calculation unit 250 uses the decision table having the attribute (for example, the recommendation algorithm score) used for generating the content list by the content list generation unit 240 as the determination attribute. It may be calculated. Similar to the importance calculation unit 150 described above, the importance calculation unit 250 outputs the importance calculated for each of the plurality of condition attribute sets to the extraction unit 260.

  The extraction unit 260 extracts one or more important attributes used to generate the title of the content list according to the above-described importance calculated by the importance calculation unit 250. Similar to the extraction unit 160 described above, the important attribute extraction processing by the extraction unit 260 can also be performed by two patterns of direct extraction and stepwise extraction.

  According to such a modification, for example, even when a content list is generated in accordance with a user's specification, there are combinations of important attributes related to the user's intention or emotion that the system cannot directly know. The title of the content list can be dynamically generated using the extracted important attributes. Further, for example, when the system generates a content list using a recommendation algorithm, a title suitable for the result of the recommendation can be dynamically generated. The title of the content list may be, for example, a simple combination of attribute character strings such as “Rock, Scandal”, or may be processed more like a title instead.

[4-2. Personalization]
FIG. 18 is an explanatory diagram for explaining another modified example of the present embodiment. In the upper part of FIG. 18, an attribute table held by the storage unit 110 shown in FIG. 1 is shown. Also, below the attribute table, history data representing the history of user actions for each content of the user U1 is shown. Such history data is held by the storage unit 110 for each user (or for each device), for example. For example, the importance level calculation unit 150 or 250 acquires a subset of the attribute table for the content included in the history data, and calculates the importance level based only on the attribute value included in the acquired subset. In the example of FIG. 18, user actions for the contents C1, C4, and C2 are described in the history data of the user U1, and a subset of the attribute table for the contents C1, C2, and C4 is acquired. In such a case, the important attribute output from the extraction unit 160 or 260 is an important attribute for a specific user (user U1 in the example of FIG. 18), not all users. By filtering the attribute table based on such history data, it is possible to calculate the importance according to the characteristics of the individual user. As a result, it is possible to achieve personalization of the reason for recommendation presented by the information processing apparatus 100 or 200 or the title of the playlist, and the user satisfaction can be improved.

<5. Hardware configuration example>
A series of processing by the information processing apparatus 100 or 200 described above is typically realized using software. The program constituting the software is executed using, for example, a general-purpose computer having the configuration shown in FIG.

  In FIG. 19, a CPU (Central Processing Unit) 902 controls the overall operation of the general-purpose computer. A ROM (Read Only Memory) 904 stores a program or data describing a part or all of a series of processes. A RAM (Random Access Memory) 906 temporarily stores programs and data used by the CPU 902 when processing is executed.

  The CPU 902, ROM 904, and RAM 906 are connected to each other via a bus 910. An input / output interface 912 is further connected to the bus 910.

  The input / output interface 912 is an interface for connecting the CPU 902, ROM 904, and RAM 906 to the input device 920, output device 922, storage device 924, communication device 926, and drive 930.

  The input device 920 receives an instruction and information input from the user via an input unit such as a mouse, a keyboard, or a touch panel. The output device 922 outputs information to the user via a display device such as a CRT (Cathode Ray Tube), a liquid crystal display, an OLED (Organic Light Emitting Diode), or an audio output device such as a speaker.

  The storage device 924 is configured by, for example, a hard disk or a flash memory, and stores programs, program data, and the like. The communication device 926 performs communication processing via a network such as a LAN or the Internet. The drive 930 is provided in a general-purpose computer as necessary. For example, a removable medium 932 is attached to the drive 930.

<6. Summary>
Up to this point, an embodiment of the present invention and its modifications have been described with reference to FIGS. According to the present embodiment, one or more importance levels other than the predetermined attribute of the content are represented by using the attribute value given to each content with respect to a plurality of contents, and representing the degree of contribution to the predetermined attribute of the content. Are calculated for the attributes. At this time, the importance is calculated using a decision table in which the predetermined attribute is a decision attribute and the one or more attributes are condition attributes. As a result, the importance can be flexibly evaluated for various combinations of condition attributes.

  Further, according to the present embodiment, the importance is calculated based on the number of contents forming the positive region of the decision attribute for the decision table according to the Rough set theory. In general, the density of the positive region of the decision table can be treated as an index of information transferability with respect to the attribute value of the decision attribute of the combination of condition attributes. Therefore, the importance calculated by the method according to the present embodiment is a significant index corresponding to information transmissibility for each combination of condition attributes.

  Further, according to the present embodiment, information to be presented to the user regarding the content, for example, the reason for recommendation or the title of the content list, is generated using the important attribute extracted according to the calculated importance. Thereby, even when a combination of mutually related attributes is meaningful, the combination of the attributes can be accurately reflected in a recommendation reason or a title of a content list. In addition, the reason for recommendation or the title of the content list can be adaptively changed in accordance with the situation (that is, context) at the time of user feedback or user action.

  Further, according to the present embodiment, the attribute value of the extended attribute is calculated based on the attribute value of the basic attribute according to the probabilistic classification method by PLSA or LDA. And an important attribute can be extracted in steps by using the attribute value of an extended attribute in the middle. As a result, it is possible to eliminate the risk that the accuracy of the calculation result of the degree of importance is reduced due to the shake of the notation of the basic attribute (for example, keyword or person name). In addition, it is possible to generate a recommendation reason or the like after properly considering a potential relationship such as a synonym relationship between basic attributes.

  Note that according to the importance calculation method using the determination table described in this specification, various types of importance can be calculated according to the setting of the condition attribute and the determination attribute. For example, by using the context data representing the user's state acquired by the above-described smile sensor or biometric sensor as a determination attribute, it is possible to calculate a different degree of importance depending on the user's psychological state or physical condition.

  The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to such examples. It is obvious that a person having ordinary knowledge in the technical field to which the present invention pertains can come up with various changes or modifications within the scope of the technical idea described in the claims. Of course, it is understood that these also belong to the technical scope of the present invention.

100, 200 Information processing device 110 Storage unit 120 Analysis unit 130 User interface control unit 140 Recommendation unit 150, 250 Importance calculation unit 160, 260 Extraction unit 240 Content list generation unit

Claims (16)

A storage unit that holds an attribute table that stores attribute values assigned to each content for a plurality of contents;
An importance calculation unit that calculates the importance of one or more other attributes with respect to a predetermined attribute of the content using the attribute value stored in the attribute table;
With
The importance calculating unit calculates the importance using a determination table in which the predetermined attribute is a determination attribute and the one or more attributes are condition attributes;
Information processing device.   The information processing apparatus according to claim 1, wherein the importance calculation unit calculates the importance based on a number of contents forming a positive region of the determination attribute for the determination table. The information processing apparatus includes:
An extraction unit that extracts one or more important attributes used to generate information about the content to be presented to the user according to the importance calculated by the importance calculation unit;
The information processing apparatus according to claim 2, further comprising: The information processing apparatus includes:
A recommendation unit that selects content to be recommended to a user using an attribute value stored in the attribute table, and generates a reason for recommendation based on the one or more important attributes extracted by the extraction unit Recommendation section,
The information processing apparatus according to claim 3, further comprising: The recommendation unit further stores a score for each content calculated for content selection in the attribute table,
The extraction unit extracts the one or more important attributes according to the importance calculated by the importance calculation unit using the score as a determination attribute.
The information processing apparatus according to claim 4. The attribute table further stores attribute values of feedback attributes given based on user feedback for each content,
The recommendation unit selects content to be recommended to the user using the attribute value of the feedback attribute,
The extraction unit extracts the one or more important attributes according to the importance calculated by the importance calculation unit using the feedback attribute as a determination attribute;
The information processing apparatus according to claim 4. The attribute table further stores an attribute value of a context attribute that is given based on a user action status for each content,
The recommendation unit selects content to be recommended to the user using the attribute value of the context attribute,
The extraction unit extracts the one or more important attributes according to the importance calculated by the importance calculation unit using the context attribute as a determination attribute;
The information processing apparatus according to claim 4.   The information processing according to claim 3, wherein the attribute table stores attribute values of extended attributes obtained by analyzing attribute values of the basic attributes in addition to attribute values of the basic attributes assigned to each content. apparatus.   The extraction unit includes the first attribute calculated by the importance calculation unit using the predetermined attribute as a determination attribute and one or more attributes included in the extended attribute as a condition attribute, and the extended attribute. The one or more important attributes are extracted according to the second importance calculated by the importance calculation unit using the attribute to be determined as the determination attribute and one or more attributes included in the basic attribute as the condition attributes The information processing apparatus according to claim 8. The information processing apparatus includes:
In accordance with a probabilistic classification method based on PLSA (Probabilistic Latent Semantic Analysis) or LDA (Latent Dirichlet Allocation), an analysis unit that calculates the attribute value of the extended attribute based on the attribute value of the basic attribute;
The information processing apparatus according to claim 9, further comprising: The information processing apparatus includes:
A content list generation unit that generates a list of contents to be played back using attribute values stored in the attribute table, wherein the content list is generated based on the one or more important attributes extracted by the extraction unit. A content list generator for generating titles;
Further comprising
The extraction unit includes the one or more important attributes according to the importance calculated by the importance calculation unit using the attribute used for generating the content list by the content list generation unit as a determination attribute. Extract,
The information processing apparatus according to claim 3. The information processing apparatus includes:
A content list generation unit that generates a list of contents to be played according to a user's specification, wherein the content list generation unit generates a title of the content list based on the one or more important attributes extracted by the extraction unit Part,
Further comprising
The extraction unit extracts the one or more important attributes according to the importance calculated by the importance calculation unit using an attribute whose attribute value is determined according to whether or not specified by the user as a determination attribute. ,
The information processing apparatus according to claim 3. The storage unit further holds history data representing a history of user actions for each content,
The importance level calculation unit calculates the importance level for the one or more attributes for each user using an attribute value for content included in the history data.
The information processing apparatus according to claim 1.   The information processing apparatus according to claim 1, wherein the importance calculation unit derives a positive region of a decision attribute in the decision table according to a Rough set theory. In an information processing apparatus that uses a storage medium to store an attribute table that stores attribute values assigned to each content for a plurality of contents,
Calculating the importance of one or more other attributes with respect to a predetermined attribute of the content using the attribute values stored in the attribute table;
An importance calculation method including
The importance is calculated using a determination table in which the predetermined attribute is a determination attribute and the one or more attributes are condition attributes.
Importance calculation method. For a plurality of contents, a computer that controls an information processing apparatus that holds an attribute table that stores attribute values assigned to each content using a storage medium,
An importance calculation unit that calculates the importance of one or more other attributes with respect to a predetermined attribute of the content using the attribute value stored in the attribute table;
Is a program for functioning as
The importance calculating unit calculates the importance using a determination table in which the predetermined attribute is a determination attribute and the one or more attributes are condition attributes;
program.

Images